Making Your Devices Smart - Stanford University

Making Your Devices Smart

Guha Jayachandran @guha guha@onutechnology.com

October 15, 2015

@guha

Outline

Opportunity and challenge

Dream

Options

Examples

Invitation

@guha

Revolution

@guha

Today’s Advances are Real but Hype is Not New

2014 Gartner Hype Cycle

“A.I. Winters” followed previous

waves of excitement

@guha

Computational power

Data Algorithms

MagicBut…

@guha

Why can’t we make that sort of

magic ourselves?

@guha

What has allowed anyone to do it?

@guha

Enablers of Today’s Revolution: Unprecedented Computation, Unprecedented Data…

1965 Electronics magazine From: Fairchild internal document

@guha

Data, data everywhere…

~5 Exabytes

All of human history

until 2004

~5 Exabytes

Today

12:00AM

11:59PM

@guha

Computational power

Data Algorithms

Magic

@guha

Exploit Parallelism

@guha

Exploit Parallelism

@guha

Exploit Parallelism

Facebook data center (via AnandTech:http://www.anandtech.com/show/4531/ask-the-experts-enterprise-cloud-computing-questions-answered-part-1

@guha

Why can’t we make that sort of

magic ourselves?

@guha

You need:

1. A capable computational technique

2. That technique to be quick and accessible

@guha

The Problem in ShortProgramming all but the simplest algorithms is hard and slow!

Even well-understood algorithms take substantial allocation of time and developer resources to an area that is immensely critical but not unique to your business.

New parallel architectures are difficult to exploit.

And it gets really hard and expensive if you want to run algorithms on big data with instant performance.

Data streams today are easily TB to PB, and at least require low latency. Even GB-scale can require data center or cloud infrastructure, installation/maintenance of packages, Hadoop, UI for analysts, and on and on. Even if you’re willing to pay the above costs, the expertise required for high-performance code is well beyond the capabilities of the vast majority of companies, large or small.

@guha

Why can’t we make that sort of

magic ourselves?

@guha

Can’t others enable us to make that

sort of magic ourselves?

Yes!

@guha

APIs

@guha

Onu: Enabling IoTWith just a couple lines of code to tap into our system,

an app can recognize objects in a live stream

an image set or video can be indexed or queried

a transcript of a live conversation can be produced

…

plus algorithms and applications outside IoT

@guha

Solution AttributesEasy to use

Seamless for engineers to leverage Intuitive interfaces

Powerful Scalably handles any size data Rich suite of algorithms Rely on our algorithmic and process optimization expertise

Fast Very highly optimized and parallelized code; low latency, high throughput Makes use of the best architecture for the problem, including GPUs, SSDs, etc., and the best ordering (“Cloud Compiler”)

Flexible Use hosted-cloud solution or a system that runs on your local infrastructure or code for your device Data can live where you want

Open No proprietary workflow trap because it plays nice; fully transparent Data belongs to you

@guha

Want to try the system? Send us a mail to be notified for the next batch of invitations.

anticipate@onutechnology.com

@guha

Image Work OngoingSteady progress

@guha

But not yet perfect

@guha

@guha

@guha

@guha

@guha

Our quadcopter with a smartphone attached!

@guha

–Alan Turing

“We can only see a short distance ahead, but we can see plenty there

that needs to be done.”

@guha

Onu TechnologyThe world’s most powerful algorithms, for everyone

Get in touch if you’re interested in working together:info@onutechnology.com

Sign up for waiting list:anticipate@onutechnology.com